idnits 2.17.1 draft-ietf-httpbis-priority-09.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- No issues found here. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (10 November 2021) is 897 days in the past. Is this intentional? -- Found something which looks like a code comment -- if you have code sections in the document, please surround them with '' and '' lines. Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) == Missing Reference: '-1' is mentioned on line 1271, but not defined -- Looks like a reference, but probably isn't: '6' on line 1271 -- Looks like a reference, but probably isn't: '0' on line 1271 -- Looks like a reference, but probably isn't: '7' on line 1271 -- Possible downref: Normative reference to a draft: ref. 'HTTP' == Outdated reference: A later version (-07) exists of draft-ietf-httpbis-http2bis-05 -- Obsolete informational reference (is this intentional?): RFC 7540 (Obsoleted by RFC 9113) Summary: 0 errors (**), 0 flaws (~~), 3 warnings (==), 7 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 HTTP K. Oku 3 Internet-Draft Fastly 4 Intended status: Standards Track L. Pardue 5 Expires: 14 May 2022 Cloudflare 6 10 November 2021 8 Extensible Prioritization Scheme for HTTP 9 draft-ietf-httpbis-priority-09 11 Abstract 13 This document describes a scheme that allows an HTTP client to 14 communicate its preferences for how the upstream server prioritizes 15 responses to its requests, and also allows a server to hint to a 16 downstream intermediary how its responses should be prioritized when 17 they are forwarded. This document defines the Priority header field 18 for communicating the initial priority in an HTTP version-independent 19 manner, as well as HTTP/2 and HTTP/3 frames for reprioritizing 20 responses. These share a common format structure that is designed to 21 provide future extensibility. 23 Note to Readers 25 _RFC EDITOR: please remove this section before publication_ 27 Discussion of this draft takes place on the HTTP working group 28 mailing list (ietf-http-wg@w3.org), which is archived at 29 https://lists.w3.org/Archives/Public/ietf-http-wg/ 30 (https://lists.w3.org/Archives/Public/ietf-http-wg/). 32 Working Group information can be found at https://httpwg.org/ 33 (https://httpwg.org/); source code and issues list for this draft can 34 be found at https://github.com/httpwg/http-extensions/labels/ 35 priorities (https://github.com/httpwg/http-extensions/labels/ 36 priorities). 38 Status of This Memo 40 This Internet-Draft is submitted in full conformance with the 41 provisions of BCP 78 and BCP 79. 43 Internet-Drafts are working documents of the Internet Engineering 44 Task Force (IETF). Note that other groups may also distribute 45 working documents as Internet-Drafts. The list of current Internet- 46 Drafts is at https://datatracker.ietf.org/drafts/current/. 48 Internet-Drafts are draft documents valid for a maximum of six months 49 and may be updated, replaced, or obsoleted by other documents at any 50 time. It is inappropriate to use Internet-Drafts as reference 51 material or to cite them other than as "work in progress." 53 This Internet-Draft will expire on 14 May 2022. 55 Copyright Notice 57 Copyright (c) 2021 IETF Trust and the persons identified as the 58 document authors. All rights reserved. 60 This document is subject to BCP 78 and the IETF Trust's Legal 61 Provisions Relating to IETF Documents (https://trustee.ietf.org/ 62 license-info) in effect on the date of publication of this document. 63 Please review these documents carefully, as they describe your rights 64 and restrictions with respect to this document. Code Components 65 extracted from this document must include Simplified BSD License text 66 as described in Section 4.e of the Trust Legal Provisions and are 67 provided without warranty as described in the Simplified BSD License. 69 Table of Contents 71 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 72 1.1. Notational Conventions . . . . . . . . . . . . . . . . . 4 73 2. Motivation for Replacing RFC 7540 Priorities . . . . . . . . 5 74 2.1. Disabling RFC 7540 Priorities . . . . . . . . . . . . . . 6 75 2.1.1. Advice when Using Extensible Priorities as the 76 Alternative . . . . . . . . . . . . . . . . . . . . . 7 77 3. Applicability of the Extensible Priority Scheme . . . . . . . 8 78 4. Priority Parameters . . . . . . . . . . . . . . . . . . . . . 8 79 4.1. Urgency . . . . . . . . . . . . . . . . . . . . . . . . . 9 80 4.2. Incremental . . . . . . . . . . . . . . . . . . . . . . . 9 81 4.3. Defining New Parameters . . . . . . . . . . . . . . . . . 10 82 4.3.1. Registration . . . . . . . . . . . . . . . . . . . . 11 83 5. The Priority HTTP Header Field . . . . . . . . . . . . . . . 12 84 6. Reprioritization . . . . . . . . . . . . . . . . . . . . . . 12 85 7. The PRIORITY_UPDATE Frame . . . . . . . . . . . . . . . . . . 12 86 7.1. HTTP/2 PRIORITY_UPDATE Frame . . . . . . . . . . . . . . 13 87 7.2. HTTP/3 PRIORITY_UPDATE Frame . . . . . . . . . . . . . . 15 88 8. Merging Client- and Server-Driven Parameters . . . . . . . . 16 89 9. Client Scheduling . . . . . . . . . . . . . . . . . . . . . . 17 90 10. Server Scheduling . . . . . . . . . . . . . . . . . . . . . . 17 91 10.1. Intermediaries with Multiple Backend Connections . . . . 19 92 11. Scheduling and the CONNECT Method . . . . . . . . . . . . . . 19 93 12. Retransmission Scheduling . . . . . . . . . . . . . . . . . . 20 94 13. Fairness . . . . . . . . . . . . . . . . . . . . . . . . . . 20 95 13.1. Coalescing Intermediaries . . . . . . . . . . . . . . . 20 96 13.2. HTTP/1.x Back Ends . . . . . . . . . . . . . . . . . . . 21 97 13.3. Intentional Introduction of Unfairness . . . . . . . . . 21 98 14. Why use an End-to-End Header Field? . . . . . . . . . . . . . 22 99 15. Security Considerations . . . . . . . . . . . . . . . . . . . 22 100 16. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 23 101 17. References . . . . . . . . . . . . . . . . . . . . . . . . . 23 102 17.1. Normative References . . . . . . . . . . . . . . . . . . 24 103 17.2. Informative References . . . . . . . . . . . . . . . . . 24 104 Appendix A. Acknowledgements . . . . . . . . . . . . . . . . . . 25 105 Appendix B. Change Log . . . . . . . . . . . . . . . . . . . . . 26 106 B.1. Since draft-ietf-httpbis-priority-08 . . . . . . . . . . 26 107 B.2. Since draft-ietf-httpbis-priority-07 . . . . . . . . . . 26 108 B.3. Since draft-ietf-httpbis-priority-06 . . . . . . . . . . 26 109 B.4. Since draft-ietf-httpbis-priority-05 . . . . . . . . . . 26 110 B.5. Since draft-ietf-httpbis-priority-04 . . . . . . . . . . 27 111 B.6. Since draft-ietf-httpbis-priority-03 . . . . . . . . . . 27 112 B.7. Since draft-ietf-httpbis-priority-02 . . . . . . . . . . 27 113 B.8. Since draft-ietf-httpbis-priority-01 . . . . . . . . . . 27 114 B.9. Since draft-ietf-httpbis-priority-00 . . . . . . . . . . 27 115 B.10. Since draft-kazuho-httpbis-priority-04 . . . . . . . . . 28 116 B.11. Since draft-kazuho-httpbis-priority-03 . . . . . . . . . 28 117 B.12. Since draft-kazuho-httpbis-priority-02 . . . . . . . . . 28 118 B.13. Since draft-kazuho-httpbis-priority-01 . . . . . . . . . 28 119 B.14. Since draft-kazuho-httpbis-priority-00 . . . . . . . . . 28 120 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 29 122 1. Introduction 124 It is common for representations of an HTTP [HTTP] resource to have 125 relationships to one or more other resources. Clients will often 126 discover these relationships while processing a retrieved 127 representation, which may lead to further retrieval requests. 128 Meanwhile, the nature of the relationship determines whether the 129 client is blocked from continuing to process locally available 130 resources. An example of this is visual rendering of an HTML 131 document, which could be blocked by the retrieval of a CSS file that 132 the document refers to. In contrast, inline images do not block 133 rendering and get drawn incrementally as the chunks of the images 134 arrive. 136 HTTP/2 [HTTP2] and HTTP/3 [HTTP3] support multiplexing of requests 137 and responses in a single connection. An important feature of any 138 implementation of a protocol that provides multiplexing is the 139 ability to prioritize the sending of information. For example, to 140 provide meaningful presentation of an HTML document at the earliest 141 moment, it is important for an HTTP server to prioritize the HTTP 142 responses, or the chunks of those HTTP responses, that it sends to a 143 client. 145 A server that operates in ignorance of how clients issue requests and 146 consume responses can cause suboptimal client application 147 performance. Priority signals allow clients to communicate their 148 view of request priority. Servers have their own needs that are 149 independent from client needs, so they often combine priority signals 150 with other available information in order to inform scheduling of 151 response data. 153 RFC 7540 [RFC7540] stream priority allowed a client to send a series 154 of priority signals that communicate to the server a "priority tree"; 155 the structure of this tree represents the client's preferred relative 156 ordering and weighted distribution of the bandwidth among HTTP 157 responses. Servers could use these priority signals as input into 158 prioritization decision making. 160 The design and implementation of RFC 7540 stream priority was 161 observed to have shortcomings, explained in Section 2. HTTP/2 162 [HTTP2] has consequently deprecated the use of these stream priority 163 signals. 165 This document describes an extensible scheme for prioritizing HTTP 166 responses that uses absolute values. Section 4 defines priority 167 parameters, which are a standardized and extensible format of 168 priority information. Section 5 defines the Priority HTTP header 169 field, a protocol-version-independent and end-to-end priority signal. 170 Clients can use this header to signal priority to servers in order to 171 specify the precedence of HTTP responses. Similarly, servers behind 172 an intermediary can use it to signal priority to the intermediary. 173 Section 7.1 and Section 7.2 define version-specific frames that carry 174 parameters, which clients can use for reprioritization. 176 Header field and frame priority signals are input to a server's 177 response prioritization process. They are only a suggestion and do 178 not guarantee any particular processing or transmission order for one 179 response relative to any other response. Section 10 and Section 12 180 provide consideration and guidance about how servers might act upon 181 signals. 183 The prioritization scheme and priority signals defined herein can act 184 as a substitute for RFC 7540 stream priority. 186 1.1. Notational Conventions 188 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 189 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 190 document are to be interpreted as described in [RFC2119]. 192 The terms Dictionary, sf-boolean, sf-dictionary, and sf-integer are 193 imported from [STRUCTURED-FIELDS]. 195 Example HTTP requests and responses use the HTTP/2-style formatting 196 from [HTTP2]. 198 This document uses the variable-length integer encoding from [QUIC]. 200 The term control stream is used to describe the HTTP/2 stream with 201 identifier 0x0, and HTTP/3 control stream; see Section 6.2.1 of 202 [HTTP3]. 204 The term HTTP/2 priority signal is used to describe the priority 205 information sent from clients to servers in HTTP/2 frames; see 206 Section 5.3.2 of [HTTP2]. 208 2. Motivation for Replacing RFC 7540 Priorities 210 RFC 7540 stream priority (see Section 5.3 of [RFC7540]) is a complex 211 system where clients signal stream dependencies and weights to 212 describe an unbalanced tree. It suffered from limited deployment and 213 interoperability and was deprecated in a revision of HTTP/2 [HTTP2]. 214 However, in order to maintain wire compatibility, HTTP/2 priority 215 signals are still mandatory to handle (see Section 5.3.2 of [HTTP2]). 217 Clients can build RFC 7540 trees with rich flexibility but experience 218 has shown this is rarely exercised. Instead they tend to choose a 219 single model optimized for a single use case and experiment within 220 the model constraints, or do nothing at all. Furthermore, many 221 clients build their prioritization tree in a unique way, which makes 222 it difficult for servers to understand their intent and act or 223 intervene accordingly. 225 Many RFC 7540 server implementations do not act on HTTP/2 priority 226 signals. Some instead favor custom server-driven schemes based on 227 heuristics or other hints, such as resource content type or request 228 generation order. For example, a server, with knowledge of an HTML 229 document structure, might want to prioritize the delivery of images 230 that are critical to user experience above other images, but below 231 the CSS files. Since client trees vary, it is impossible for the 232 server to determine how such images should be prioritized against 233 other responses. 235 RFC 7540 allows intermediaries to coalesce multiple client trees into 236 a single tree that is used for a single upstream HTTP/2 connection. 237 However, most intermediaries do not support this. Additionally, RFC 238 7540 does not define a method that can be used by a server to express 239 the priority of a response. Without such a method, intermediaries 240 cannot coordinate client-driven and server-driven priorities. 242 RFC 7540 describes denial-of-service considerations for 243 implementations. On 2019-08-13 Netflix issued an advisory notice 244 about the discovery of several resource exhaustion vectors affecting 245 multiple RFC 7540 implementations. One attack, [CVE-2019-9513] aka 246 "Resource Loop", is based on using priority signals to manipulate the 247 server's stored prioritization state. 249 HTTP/2 priority associated with an HTTP request is signalled as a 250 value relative to those of other requests sharing the same HTTP/2 251 connection. Therefore, in order to prioritize requests, endpoints 252 are compelled to have the knowledge of the underlying HTTP version 253 and how the requests are coalesced. This has been a burden to HTTP 254 endpoints that generate or forward requests in a version-agnostic 255 manner. 257 HTTP/2 priority signals are required to be delivered and processed in 258 the order they are sent so that the receiver handling is 259 deterministic. Porting HTTP/2 priority signals to protocols that do 260 not provide ordering guarantees presents challenges. For example, 261 HTTP/3 [HTTP3] lacks global ordering across streams that would carry 262 priority signals. Early attempts to port HTTP/2 priority signals to 263 HTTP/3 required adding additional information to the signals, leading 264 to more complicated processing. Problems found with this approach 265 could not be resolved and definition of a HTTP/3 priority signalling 266 feature was removed before publication. 268 Considering the deployment problems and the design restrictions of 269 RFC 7540 stream priority, as well as the difficulties in adapting it 270 to HTTP/3, continuing to base prioritization on this mechanism risks 271 increasing the complexity of systems. Multiple experiments from 272 independent research have shown that simpler schemes can reach at 273 least equivalent performance characteristics compared to the more 274 complex RFC 7540 setups seen in practice, at least for the web use 275 case. 277 2.1. Disabling RFC 7540 Priorities 279 The problems and insights set out above provided the motivation for 280 deprecating RFC 7540 stream priority (see Section 5.3 of [RFC7540]). 282 The SETTINGS_NO_RFC7540_PRIORITIES HTTP/2 setting is defined by this 283 document in order to allow endpoints to omit or ignore HTTP/2 284 priority signals (see Section 5.3.2 of [HTTP2]), as described below. 285 The value of SETTINGS_NO_RFC7540_PRIORITIES MUST be 0 or 1. Any 286 value other than 0 or 1 MUST be treated as a connection error (see 287 Section 5.4.1 of [HTTP2]) of type PROTOCOL_ERROR. The initial value 288 is 0. 290 If endpoints use SETTINGS_NO_RFC7540_PRIORITIES they MUST send it in 291 the first SETTINGS frame. Senders MUST NOT change the 292 SETTINGS_NO_RFC7540_PRIORITIES value after the first SETTINGS frame. 293 Receivers that detect a change MAY treat it as a connection error of 294 type PROTOCOL_ERROR. 296 Clients can send SETTINGS_NO_RFC7540_PRIORITIES with a value of 1 to 297 indicate that they are not using HTTP/2 priority signals. The 298 SETTINGS frame precedes any HTTP/2 priority signal sent from clients, 299 so servers can determine whether they need to allocate any resources 300 to signal handling before signals arrive. A server that receives 301 SETTINGS_NO_RFC7540_PRIORITIES with a value of 1 MUST ignore HTTP/2 302 priority signals. 304 Servers can send SETTINGS_NO_RFC7540_PRIORITIES with a value of 1 to 305 indicate that they will ignore HTTP/2 priority signals sent by 306 clients. 308 Endpoints that send SETTINGS_NO_RFC7540_PRIORITIES are encouraged to 309 use alternative priority signals (for example, Section 5 or 310 Section 7.1) but there is no requirement to use a specific signal 311 type. 313 2.1.1. Advice when Using Extensible Priorities as the Alternative 315 Until the client receives the SETTINGS frame from the server, the 316 client SHOULD send both the HTTP/2 priority signals and the signals 317 of this prioritization scheme (see Section 5 and Section 7.1). When 318 the client receives the first SETTINGS frame that contains the 319 SETTINGS_NO_RFC7540_PRIORITIES parameter with value of 1, it SHOULD 320 stop sending the HTTP/2 priority signals. If the value was 0 or if 321 the settings parameter was absent, it SHOULD stop sending 322 PRIORITY_UPDATE frames (Section 7.1), but MAY continue sending the 323 Priority header field (Section 5), as it is an end-to-end signal that 324 might be useful to nodes behind the server that the client is 325 directly connected to. 327 3. Applicability of the Extensible Priority Scheme 329 The priority scheme defined by this document considers only the 330 prioritization of HTTP messages and tunnels, see Section 9, 331 Section 10, and Section 11. 333 Where HTTP extensions change stream behavior or define new data 334 carriage mechanisms, they can also define how this priority scheme 335 can be applied. 337 4. Priority Parameters 339 The priority information is a sequence of key-value pairs, providing 340 room for future extensions. Each key-value pair represents a 341 priority parameter. 343 The Priority HTTP header field (Section 5) is an end-to-end way to 344 transmit this set of parameters when a request or a response is 345 issued. In order to reprioritize a request, HTTP-version-specific 346 PRIORITY_UPDATE frames (Section 7.1 and Section 7.2) are used by 347 clients to transmit the same information on a single hop. 349 Intermediaries can consume and produce priority signals in a 350 PRIORITY_UPDATE frame or Priority header field. Sending a 351 PRIORITY_UPDATE frame preserves the signal from the client, but 352 provides a signal that overrides that for the next hop; see 353 Section 14. Replacing or adding a Priority header field overrides 354 any signal from a client and can affect prioritization for all 355 subsequent recipients. 357 For both the Priority header field and the PRIORITY_UPDATE frame, the 358 set of priority parameters is encoded as a Structured Fields 359 Dictionary (see Section 3.2 of [STRUCTURED-FIELDS]). 361 This document defines the urgency(u) and incremental(i) parameters. 362 When receiving an HTTP request that does not carry these priority 363 parameters, a server SHOULD act as if their default values were 364 specified. Note that handling of omitted parameters is different 365 when processing an HTTP response; see Section 8. 367 Receivers parse the Dictionary as defined in Section 4.2 of 368 [STRUCTURED-FIELDS]. Where the Dictionary is successfully parsed, 369 this document places the additional requirement that unknown priority 370 parameters, parameters with out-of-range values, or values of 371 unexpected types MUST be ignored. 373 4.1. Urgency 375 The urgency parameter (u) takes an integer between 0 and 7, in 376 descending order of priority. 378 The value is encoded as an sf-integer. The default value is 3. 380 Endpoints use this parameter to communicate their view of the 381 precedence of HTTP responses. The chosen value of urgency can be 382 based on the expectation that servers might use this information to 383 transmit HTTP responses in the order of their urgency. The smaller 384 the value, the higher the precedence. 386 The following example shows a request for a CSS file with the urgency 387 set to 0: 389 :method = GET 390 :scheme = https 391 :authority = example.net 392 :path = /style.css 393 priority = u=0 395 A client that fetches a document that likely consists of multiple 396 HTTP resources (e.g., HTML) SHOULD assign the default urgency level 397 to the main resource. This convention allows servers to refine the 398 urgency using knowledge specific to the web-site (see Section 8). 400 The lowest urgency level (7) is reserved for background tasks such as 401 delivery of software updates. This urgency level SHOULD NOT be used 402 for fetching responses that have impact on user interaction. 404 4.2. Incremental 406 The incremental parameter (i) takes an sf-boolean as the value that 407 indicates if an HTTP response can be processed incrementally, i.e., 408 provide some meaningful output as chunks of the response arrive. 410 The default value of the incremental parameter is false (0). 412 If a client makes concurrent requests with the incremental parameter 413 set to false, there is no benefit serving responses with the same 414 urgency concurrently because the client is not going to process those 415 responses incrementally. Serving non-incremental responses with the 416 same urgency one by one, in the order in which those requests were 417 generated is considered to be the best strategy. 419 If a client makes concurrent requests with the incremental parameter 420 set to true, serving requests with the same urgency concurrently 421 might be beneficial. Doing this distributes the connection 422 bandwidth, meaning that responses take longer to complete. 423 Incremental delivery is most useful where multiple partial responses 424 might provide some value to clients ahead of a complete response 425 being available. 427 The following example shows a request for a JPEG file with the 428 urgency parameter set to 5 and the incremental parameter set to true. 430 :method = GET 431 :scheme = https 432 :authority = example.net 433 :path = /image.jpg 434 priority = u=5, i 436 4.3. Defining New Parameters 438 When attempting to define new parameters, care must be taken so that 439 they do not adversely interfere with prioritization performed by 440 existing endpoints or intermediaries that do not understand the newly 441 defined parameter. Since unknown parameters are ignored, new 442 parameters should not change the interpretation of, or modify, the 443 urgency (see Section 4.1) or incremental (see Section 4.2) parameters 444 in a way that is not backwards compatible or fallback safe. 446 For example, if there is a need to provide more granularity than 447 eight urgency levels, it would be possible to subdivide the range 448 using an additional parameter. Implementations that do not recognize 449 the parameter can safely continue to use the less granular eight 450 levels. 452 Alternatively, the urgency can be augmented. For example, a 453 graphical user agent could send a visible parameter to indicate if 454 the resource being requested is within the viewport. 456 Generic parameters are preferred over vendor-specific, application- 457 specific or deployment-specific values. If a generic value cannot be 458 agreed upon in the community, the parameter's name should be 459 correspondingly specific (e.g., with a prefix that identifies the 460 vendor, application or deployment). 462 4.3.1. Registration 464 New Priority parameters can be defined by registering them in the 465 HTTP Priority Parameters Registry. The registry governs the keys 466 (short textual strings) used in Structured Fields Dictionary (see 467 Section 3.2 of [STRUCTURED-FIELDS]). Since each HTTP request can 468 have associated priority signals, there is value in having short key 469 lengths, especially single-character strings. In order to encourage 470 extension while avoiding unintended conflict among attractive key 471 values, the HTTP Priority Parameters Registry operates two 472 registration policies depending on key length. 474 * Registration requests for parameters with a key length of one use 475 the Specification Required policy, as per Section 4.6 of 476 [RFC8126]. 478 * Registration requests for parameters with a key length greater 479 than one use the Expert Review policy, as per Section 4.5 of 480 [RFC8126]. A specification document is appreciated, but not 481 required. 483 When reviewing registration requests, the designated expert(s) can 484 consider the additional guidance provided in Section 4.3 but cannot 485 use it as a basis for rejection. 487 Registration requests should use the following template: 489 Name: [a name for the Priority Parameter that matches key] 491 Description: [a description of the parameter semantics and value] 493 Reference: [to a specification defining this parameter] 495 See the registry at https://iana.org/assignments/http-priority 496 (https://iana.org/assignments/http-priority) for details on where to 497 send registration requests. 499 5. The Priority HTTP Header Field 501 The Priority HTTP header field carries priority parameters Section 4. 502 It can appear in requests and responses. It is an end-to-end signal 503 of the request priority from the client or the response priority from 504 the server. Section 8 describes how intermediaries can combine the 505 priority information from client requests and server responses to 506 correct or amend the precedence. Clients cannot interpret the 507 appearance or omission of a Priority response header as 508 acknowledgement that any prioritization has occurred. Guidance for 509 how endpoints can act on Priority header values is given in 510 Section 10 and Section 9. 512 Priority is a Dictionary (Section 3.2 of [STRUCTURED-FIELDS]): 514 Priority = sf-dictionary 516 As is the ordinary case for HTTP caching [CACHING], a response with a 517 Priority header field might be cached and re-used for subsequent 518 requests. When an origin server generates the Priority response 519 header field based on properties of an HTTP request it receives, the 520 server is expected to control the cacheability or the applicability 521 of the cached response, by using header fields that control the 522 caching behavior (e.g., Cache-Control, Vary). 524 6. Reprioritization 526 After a client sends a request, it may be beneficial to change the 527 priority of the response. As an example, a web browser might issue a 528 prefetch request for a JavaScript file with the urgency parameter of 529 the Priority request header field set to u=7 (background). Then, 530 when the user navigates to a page which references the new JavaScript 531 file, while the prefetch is in progress, the browser would send a 532 reprioritization signal with the priority field value set to u=0. 533 The PRIORITY_UPDATE frame (Section 7) can be used for such 534 reprioritization. 536 7. The PRIORITY_UPDATE Frame 538 This document specifies a new PRIORITY_UPDATE frame for HTTP/2 539 [HTTP2] and HTTP/3 [HTTP3]. It carries priority parameters and 540 references the target of the prioritization based on a version- 541 specific identifier. In HTTP/2, this identifier is the Stream ID; in 542 HTTP/3, the identifier is either the Stream ID or Push ID. Unlike 543 the Priority header field, the PRIORITY_UPDATE frame is a hop-by-hop 544 signal. 546 PRIORITY_UPDATE frames are sent by clients on the control stream, 547 allowing them to be sent independent from the stream that carries the 548 response. This means they can be used to reprioritize a response or 549 a push stream; or signal the initial priority of a response instead 550 of the Priority header field. 552 A PRIORITY_UPDATE frame communicates a complete set of all parameters 553 in the Priority Field Value field. Omitting a parameter is a signal 554 to use the parameter's default value. Failure to parse the Priority 555 Field Value MAY be treated as a connection error. In HTTP/2 the 556 error is of type PROTOCOL_ERROR; in HTTP/3 the error is of type 557 H3_GENERAL_PROTOCOL_ERROR. 559 A client MAY send a PRIORITY_UPDATE frame before the stream that it 560 references is open (except for HTTP/2 push streams; see Section 7.1). 561 Furthermore, HTTP/3 offers no guaranteed ordering across streams, 562 which could cause the frame to be received earlier than intended. 563 Either case leads to a race condition where a server receives a 564 PRIORITY_UPDATE frame that references a request stream that is yet to 565 be opened. To solve this condition, for the purposes of scheduling, 566 the most recently received PRIORITY_UPDATE frame can be considered as 567 the most up-to-date information that overrides any other signal. 568 Servers SHOULD buffer the most recently received PRIORITY_UPDATE 569 frame and apply it once the referenced stream is opened. Holding 570 PRIORITY_UPDATE frames for each stream requires server resources, 571 which can can be bound by local implementation policy. Although 572 there is no limit to the number of PRIORITY_UPDATES that can be sent, 573 storing only the most recently received frame limits resource 574 commitment. 576 7.1. HTTP/2 PRIORITY_UPDATE Frame 578 The HTTP/2 PRIORITY_UPDATE frame (type=0x10) is used by clients to 579 signal the initial priority of a response, or to reprioritize a 580 response or push stream. It carries the stream ID of the response 581 and the priority in ASCII text, using the same representation as the 582 Priority header field value. 584 The Stream Identifier field (see Section 5.1.1 of [HTTP2]) in the 585 PRIORITY_UPDATE frame header MUST be zero (0x0). Receiving a 586 PRIORITY_UPDATE frame with a field of any other value MUST be treated 587 as a connection error of type PROTOCOL_ERROR. 589 HTTP/2 PRIORITY_UPDATE Frame { 590 Length (24), 591 Type (i) = 10, 593 Unused Flags (8). 595 Reserved (1), 596 Stream Identifier (31), 598 Reserved (1), 599 Prioritized Stream ID (31), 600 Priority Field Value (..), 601 } 603 Figure 1: HTTP/2 PRIORITY_UPDATE Frame Payload 605 The Length, Type, Unused Flag(s), Reserved, and Stream Identifier 606 fields are described in Section 4 of [HTTP2]. The frame payload of 607 PRIORITY_UPDATE frame payload contains the following additional 608 fields: 610 Reserved: A reserved 1-bit field. The semantics of this bit are 611 undefined, and the bit MUST remain unset (0x0) when sending and 612 MUST be ignored when receiving. 614 Prioritized Stream ID: A 31-bit stream identifier for the stream 615 that is the target of the priority update. 617 Priority Field Value: The priority update value in ASCII text, 618 encoded using Structured Fields. This is the same representation 619 as the Priority header field value. 621 When the PRIORITY_UPDATE frame applies to a request stream, clients 622 SHOULD provide a Prioritized Stream ID that refers to a stream in the 623 "open", "half-closed (local)", or "idle" state. Servers can discard 624 frames where the Prioritized Stream ID refers to a stream in the 625 "half-closed (local)" or "closed" state. The number of streams which 626 have been prioritized but remain in the "idle" state plus the number 627 of active streams (those in the "open" or either "half-closed" state; 628 see Section 5.1.2 of [HTTP2]) MUST NOT exceed the value of the 629 SETTINGS_MAX_CONCURRENT_STREAMS parameter. Servers that receive such 630 a PRIORITY_UPDATE MUST respond with a connection error of type 631 PROTOCOL_ERROR. 633 When the PRIORITY_UPDATE frame applies to a push stream, clients 634 SHOULD provide a Prioritized Stream ID that refers to a stream in the 635 "reserved (remote)" or "half-closed (local)" state. Servers can 636 discard frames where the Prioritized Stream ID refers to a stream in 637 the "closed" state. Clients MUST NOT provide a Prioritized Stream ID 638 that refers to a push stream in the "idle" state. Servers that 639 receive a PRIORITY_UPDATE for a push stream in the "idle" state MUST 640 respond with a connection error of type PROTOCOL_ERROR. 642 If a PRIORITY_UPDATE frame is received with a Prioritized Stream ID 643 of 0x0, the recipient MUST respond with a connection error of type 644 PROTOCOL_ERROR. 646 If a client receives a PRIORITY_UPDATE frame, it MUST respond with a 647 connection error of type PROTOCOL_ERROR. 649 7.2. HTTP/3 PRIORITY_UPDATE Frame 651 The HTTP/3 PRIORITY_UPDATE frame (type=0xF0700 or 0xF0701) is used by 652 clients to signal the initial priority of a response, or to 653 reprioritize a response or push stream. It carries the identifier of 654 the element that is being prioritized, and the updated priority in 655 ASCII text, using the same representation as that of the Priority 656 header field value. PRIORITY_UPDATE with a frame type of 0xF0700 is 657 used for request streams, while PRIORITY_UPDATE with a frame type of 658 0xF0701 is used for push streams. 660 The PRIORITY_UPDATE frame MUST be sent on the client control stream 661 (see Section 6.2.1 of [HTTP3]). Receiving a PRIORITY_UPDATE frame on 662 a stream other than the client control stream MUST be treated as a 663 connection error of type H3_FRAME_UNEXPECTED. 665 HTTP/3 PRIORITY_UPDATE Frame { 666 Type (i) = 0xF0700..0xF0701, 667 Length (i), 668 Prioritized Element ID (i), 669 Priority Field Value (..), 670 } 672 Figure 2: HTTP/3 PRIORITY_UPDATE Frame 674 The PRIORITY_UPDATE frame payload has the following fields: 676 Prioritized Element ID: The stream ID or push ID that is the target 677 of the priority update. 679 Priority Field Value: The priority update value in ASCII text, 680 encoded using Structured Fields. This is the same representation 681 as the Priority header field value. 683 The request-stream variant of PRIORITY_UPDATE (type=0xF0700) MUST 684 reference a request stream. If a server receives a PRIORITY_UPDATE 685 (type=0xF0700) for a Stream ID that is not a request stream, this 686 MUST be treated as a connection error of type H3_ID_ERROR. The 687 Stream ID MUST be within the client-initiated bidirectional stream 688 limit. If a server receives a PRIORITY_UPDATE (type=0xF0700) with a 689 Stream ID that is beyond the stream limits, this SHOULD be treated as 690 a connection error of type H3_ID_ERROR. Generating an error is not 691 mandatory because HTTP/3 implementations might have practical 692 barriers to determining the active stream concurrency limit that is 693 applied by the QUIC layer. 695 The push-stream variant PRIORITY_UPDATE (type=0xF0701) MUST reference 696 a promised push stream. If a server receives a PRIORITY_UPDATE 697 (type=0xF0701) with a Push ID that is greater than the maximum Push 698 ID or which has not yet been promised, this MUST be treated as a 699 connection error of type H3_ID_ERROR. 701 PRIORITY_UPDATE frames of either type are only sent by clients. If a 702 client receives a PRIORITY_UPDATE frame, this MUST be treated as a 703 connection error of type H3_FRAME_UNEXPECTED. 705 8. Merging Client- and Server-Driven Parameters 707 It is not always the case that the client has the best understanding 708 of how the HTTP responses deserve to be prioritized. The server 709 might have additional information that can be combined with the 710 client's indicated priority in order to improve the prioritization of 711 the response. For example, use of an HTML document might depend 712 heavily on one of the inline images; existence of such dependencies 713 is typically best known to the server. Or, a server that receives 714 requests for a font [RFC8081] and images with the same urgency might 715 give higher precedence to the font, so that a visual client can 716 render textual information at an early moment. 718 An origin can use the Priority response header field to indicate its 719 view on how an HTTP response should be prioritized. An intermediary 720 that forwards an HTTP response can use the parameters found in the 721 Priority response header field, in combination with the client 722 Priority request header field, as input to its prioritization 723 process. No guidance is provided for merging priorities, this is 724 left as an implementation decision. 726 Absence of a priority parameter in an HTTP response indicates the 727 server's disinterest in changing the client-provided value. This is 728 different from the logic being defined for the request header field, 729 in which omission of a priority parameter implies the use of their 730 default values (see Section 4). 732 As a non-normative example, when the client sends an HTTP request 733 with the urgency parameter set to 5 and the incremental parameter set 734 to true 736 :method = GET 737 :scheme = https 738 :authority = example.net 739 :path = /menu.png 740 priority = u=5, i 742 and the origin responds with 744 :status = 200 745 content-type = image/png 746 priority = u=1 748 the intermediary might alter its understanding of the urgency from 5 749 to 1, because it prefers the server-provided value over the client's. 750 The incremental value continues to be true, the value specified by 751 the client, as the server did not specify the incremental(i) 752 parameter. 754 9. Client Scheduling 756 A client MAY use priority values to make local processing or 757 scheduling choices about the requests it initiates. 759 10. Server Scheduling 761 Priority signals are input to a prioritization process. They do not 762 guarantee any particular processing or transmission order for one 763 response relative to any other response. An endpoint cannot force a 764 peer to process concurrent request in a particular order using 765 priority. Expressing priority is therefore only a suggestion. 767 A server can use priority signals along with other inputs to make 768 scheduling decisions. No guidance is provided about how this can or 769 should be done. Factors such as implementation choices or deployment 770 environment also play a role. Any given connection is likely to have 771 many dynamic permutations. For these reasons, there is no unilateral 772 perfect scheduler and this document only provides some basic 773 recommendations for implementations. 775 Clients cannot depend on particular treatment based on priority 776 signals. Servers can use other information to prioritize responses. 778 It is RECOMMENDED that, when possible, servers respect the urgency 779 parameter (Section 4.1), sending higher urgency responses before 780 lower urgency responses. 782 The incremental parameter indicates how a client processes response 783 bytes as they arrive. It is RECOMMENDED that, when possible, servers 784 respect the incremental parameter (Section 4.2). Non-incremental 785 resources can only be used when all of the response payload has been 786 received. Therefore, non-incremental responses of the same urgency 787 SHOULD be served in their entirety, one-by-one, based on the stream 788 ID, which corresponds to the order in which clients make requests. 789 Doing so ensures that clients can use request ordering to influence 790 response order. 792 Incremental responses of the same urgency SHOULD be served by sharing 793 bandwidth amongst them. Incremental resources are used as parts, or 794 chunks, of the response payload are received. A client might benefit 795 more from receiving a portion of all these resources rather than the 796 entirety of a single resource. How large a portion of the resource 797 is needed to be useful in improving performance varies. Some 798 resource types place critical elements early, others can use 799 information progressively. This scheme provides no explicit mandate 800 about how a server should use size, type or any other input to decide 801 how to prioritize. 803 There can be scenarios where a server will need to schedule multiple 804 incremental and non-incremental responses at the same urgency level. 805 Strictly abiding the scheduling guidance based on urgency and request 806 generation order might lead to sub-optimal results at the client, as 807 early non-incremental responses might prevent serving of incremental 808 responses issued later. The following are examples of such 809 challenges. 811 1. At the same urgency level, a non-incremental request for a large 812 resource followed by an incremental request for a small resource. 814 2. At the same urgency level, an incremental request of 815 indeterminate length followed by a non-incremental large 816 resource. 818 It is RECOMMENDED that servers avoid such starvation where possible. 819 The method to do so is an implementation decision. For example, a 820 server might pre-emptively send responses of a particular incremental 821 type based on other information such as content size. 823 Optimal scheduling of server push is difficult, especially when 824 pushed resources contend with active concurrent requests. Servers 825 can consider many factors when scheduling, such as the type or size 826 of resource being pushed, the priority of the request that triggered 827 the push, the count of active concurrent responses, the priority of 828 other active concurrent responses, etc. There is no general guidance 829 on the best way to apply these. A server that is too simple could 830 easily push at too high a priority and block client requests, or push 831 at too low a priority and delay the response, negating intended goals 832 of server push. 834 Priority signals are a factor for server push scheduling. The 835 concept of parameter value defaults applies slightly differently 836 because there is no explicit client-signalled initial priority. A 837 server can apply priority signals provided in an origin response; see 838 the merging guidance given in Section 8. In the absence of origin 839 signals, applying default parameter values could be suboptimal. By 840 whatever means a server decides to schedule a pushed response, it can 841 signal the intended priority to the client by including the Priority 842 field in a PUSH_PROMISE or HEADERS frame. 844 10.1. Intermediaries with Multiple Backend Connections 846 An intermediary serving an HTTP connection might split requests over 847 multiple backend connections. When it applies prioritization rules 848 strictly, low priority requests cannot make progress while requests 849 with higher priorities are inflight. This blocking can propagate to 850 backend connections, which the peer might interpret as a connection 851 stall. Endpoints often implement protections against stalls, such as 852 abruptly closing connections after a certain time period. To reduce 853 the possibility of this occurring, intermediaries can avoid strictly 854 following prioritization and instead allocate small amounts of 855 bandwidth for all the requests that they are forwarding, so that 856 every request can make some progress over time. 858 Similarly, servers SHOULD allocate some amount of bandwidths to 859 streams acting as tunnels. 861 11. Scheduling and the CONNECT Method 863 When a request stream carries the CONNECT method, the scheduling 864 guidance in this document applies to the frames on the stream. A 865 client that issues multiple CONNECT requests can set the incremental 866 parameter to true, servers that implement the recommendation in 867 Section 10 will schedule these fairly. 869 12. Retransmission Scheduling 871 Transport protocols such as TCP and QUIC provide reliability by 872 detecting packet losses and retransmitting lost information. While 873 this document specifies HTTP-layer prioritization, its effectiveness 874 can be further enhanced if the transport layer factors priority into 875 scheduling both new data and retransmission data. The remainder of 876 this section discusses considerations when using QUIC. 878 Section 13.3 of [QUIC] states "Endpoints SHOULD prioritize 879 retransmission of data over sending new data, unless priorities 880 specified by the application indicate otherwise". When an HTTP/3 881 application uses the priority scheme defined in this document and the 882 QUIC transport implementation supports application indicated stream 883 priority, a transport that considers the relative priority of streams 884 when scheduling both new data and retransmission data might better 885 match the expectations of the application. However, there are no 886 requirements on how a transport chooses to schedule based on this 887 information because the decision depends on several factors and 888 trade-offs. It could prioritize new data for a higher urgency stream 889 over retransmission data for a lower priority stream, or it could 890 prioritize retransmission data over new data irrespective of 891 urgencies. 893 Section 6.2.4 of [QUIC-RECOVERY], also highlights consideration of 894 application priorities when sending probe packets after Probe Timeout 895 timer expiration. A QUIC implementation supporting application- 896 indicated priorities might use the relative priority of streams when 897 choosing probe data. 899 13. Fairness 901 As a general guideline, a server SHOULD NOT use priority information 902 for making scheduling decisions across multiple connections, unless 903 it knows that those connections originate from the same client. Due 904 to this, priority information conveyed over a non-coalesced HTTP 905 connection (e.g., HTTP/1.1) might go unused. 907 The remainder of this section discusses scenarios where unfairness is 908 problematic and presents possible mitigations, or where unfairness is 909 desirable. 911 13.1. Coalescing Intermediaries 913 When an intermediary coalesces HTTP requests coming from multiple 914 clients into one HTTP/2 or HTTP/3 connection going to the backend 915 server, requests that originate from one client might have higher 916 precedence than those coming from others. 918 It is sometimes beneficial for the server running behind an 919 intermediary to obey to the value of the Priority header field. As 920 an example, a resource-constrained server might defer the 921 transmission of software update files that would have the background 922 urgency being associated. However, in the worst case, the asymmetry 923 between the precedence declared by multiple clients might cause 924 responses going to one user agent to be delayed totally after those 925 going to another. 927 In order to mitigate this fairness problem, a server could use 928 knowledge about the intermediary as another signal in its 929 prioritization decisions. For instance, if a server knows the 930 intermediary is coalescing requests, then it could avoid serving the 931 responses in their entirety and instead distribute bandwidth (for 932 example, in a round-robin manner). This can work if the constrained 933 resource is network capacity between the intermediary and the user 934 agent, as the intermediary buffers responses and forwards the chunks 935 based on the prioritization scheme it implements. 937 A server can determine if a request came from an intermediary through 938 configuration, or by consulting if that request contains one of the 939 following header fields: 941 * Forwarded [FORWARDED], X-Forwarded-For 943 * Via (see Section 7.6.3 of [HTTP]) 945 13.2. HTTP/1.x Back Ends 947 It is common for CDN infrastructure to support different HTTP 948 versions on the front end and back end. For instance, the client- 949 facing edge might support HTTP/2 and HTTP/3 while communication to 950 back end servers is done using HTTP/1.1. Unlike with connection 951 coalescing, the CDN will "de-mux" requests into discrete connections 952 to the back end. HTTP/1.1 and older do not support response 953 multiplexing in a single connection, so there is not a fairness 954 problem. However, back end servers MAY still use client headers for 955 request scheduling. Back end servers SHOULD only schedule based on 956 client priority information where that information can be scoped to 957 individual end clients. Authentication and other session information 958 might provide this linkability. 960 13.3. Intentional Introduction of Unfairness 962 It is sometimes beneficial to deprioritize the transmission of one 963 connection over others, knowing that doing so introduces a certain 964 amount of unfairness between the connections and therefore between 965 the requests served on those connections. 967 For example, a server might use a scavenging congestion controller on 968 connections that only convey background priority responses such as 969 software update images. Doing so improves responsiveness of other 970 connections at the cost of delaying the delivery of updates. 972 14. Why use an End-to-End Header Field? 974 Contrary to the prioritization scheme of HTTP/2 that uses a hop-by- 975 hop frame, the Priority header field is defined as end-to-end. 977 The rationale is that the Priority header field transmits how each 978 response affects the client's processing of those responses, rather 979 than how relatively urgent each response is to others. The way a 980 client processes a response is a property associated to that client 981 generating that request. Not that of an intermediary. Therefore, it 982 is an end-to-end property. How these end-to-end properties carried 983 by the Priority header field affect the prioritization between the 984 responses that share a connection is a hop-by-hop issue. 986 Having the Priority header field defined as end-to-end is important 987 for caching intermediaries. Such intermediaries can cache the value 988 of the Priority header field along with the response, and utilize the 989 value of the cached header field when serving the cached response, 990 only because the header field is defined as end-to-end rather than 991 hop-by-hop. 993 It should also be noted that the use of a header field carrying a 994 textual value makes the prioritization scheme extensible; see the 995 discussion below. 997 15. Security Considerations 999 [RFC7540] stream prioritization relies on dependencies. 1000 Considerations are presented to implementations, describing how 1001 limiting state or work commitments can avoid some types of problems. 1002 In addition, [CVE-2019-9513] aka "Resource Loop", is an example of a 1003 DoS attack that abuses stream dependencies. Extensible priorities 1004 does not use dependencies, which avoids these issues. 1006 Section 7 describes considerations for server buffering of 1007 PRIORITY_UPDATE frames. 1009 Section 10 presents examples where servers that prioritize responses 1010 in a certain way might be starved of the ability to transmit payload. 1012 The security considerations from [STRUCTURED-FIELDS] apply to 1013 processing of priority parameters defined in Section 4. 1015 16. IANA Considerations 1017 This specification registers the following entry in the the Hypertext 1018 Transfer Protocol (HTTP) Field Name Registry established by [HTTP]: 1020 Field name: Priority 1022 Status: permanent 1024 Specification document(s): This document 1026 This specification registers the following entry in the HTTP/2 1027 Settings registry established by [RFC7540]: 1029 Name: SETTINGS_NO_RFC7540_PRIORITIES 1031 Code: 0x9 1033 Initial value: 0 1035 Specification: This document 1037 This specification registers the following entry in the HTTP/2 Frame 1038 Type registry established by [RFC7540]: 1040 Frame Type: PRIORITY_UPDATE 1042 Code: 0x10 1044 Specification: This document 1046 This specification registers the following entries in the HTTP/3 1047 Frame Type registry established by [HTTP3]: 1049 Frame Type: PRIORITY_UPDATE 1051 Code: 0xF0700 and 0xF0701 1053 Specification: This document 1055 Upon publication, please create the HTTP Priority Parameters registry 1056 at https://iana.org/assignments/http-priority 1057 (https://iana.org/assignments/http-priority) and populate it with the 1058 types defined in Section 4; see Section 4.3.1 for its associated 1059 procedures. 1061 17. References 1062 17.1. Normative References 1064 [HTTP] Fielding, R. T., Nottingham, M., and J. Reschke, "HTTP 1065 Semantics", Work in Progress, Internet-Draft, draft-ietf- 1066 httpbis-semantics-19, 12 September 2021, 1067 . 1070 [HTTP2] Thomson, M. and C. Benfield, "Hypertext Transfer Protocol 1071 Version 2 (HTTP/2)", Work in Progress, Internet-Draft, 1072 draft-ietf-httpbis-http2bis-05, 26 September 2021, 1073 . 1076 [HTTP3] Bishop, M., "Hypertext Transfer Protocol Version 3 1077 (HTTP/3)", Work in Progress, Internet-Draft, draft-ietf- 1078 quic-http-34, 2 February 2021, 1079 . 1082 [QUIC] Iyengar, J., Ed. and M. Thomson, Ed., "QUIC: A UDP-Based 1083 Multiplexed and Secure Transport", RFC 9000, 1084 DOI 10.17487/RFC9000, May 2021, 1085 . 1087 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1088 Requirement Levels", BCP 14, RFC 2119, 1089 DOI 10.17487/RFC2119, March 1997, 1090 . 1092 [RFC8126] Cotton, M., Leiba, B., and T. Narten, "Guidelines for 1093 Writing an IANA Considerations Section in RFCs", BCP 26, 1094 RFC 8126, DOI 10.17487/RFC8126, June 2017, 1095 . 1097 [STRUCTURED-FIELDS] 1098 Nottingham, M. and P-H. Kamp, "Structured Field Values for 1099 HTTP", RFC 8941, DOI 10.17487/RFC8941, February 2021, 1100 . 1102 17.2. Informative References 1104 [CACHING] Fielding, R. T., Nottingham, M., and J. Reschke, "HTTP 1105 Caching", Work in Progress, Internet-Draft, draft-ietf- 1106 httpbis-cache-19, 12 September 2021, 1107 . 1110 [CVE-2019-9513] 1111 Common Vulnerabilities and Exposures, "CVE-2019-9513", 1 1112 March 2019, . 1115 [FORWARDED] 1116 Petersson, A. and M. Nilsson, "Forwarded HTTP Extension", 1117 RFC 7239, DOI 10.17487/RFC7239, June 2014, 1118 . 1120 [I-D.lassey-priority-setting] 1121 Lassey, B. and L. Pardue, "Declaring Support for HTTP/2 1122 Priorities", Work in Progress, Internet-Draft, draft- 1123 lassey-priority-setting-00, 25 July 2019, 1124 . 1127 [QUIC-RECOVERY] 1128 Iyengar, J., Ed. and I. Swett, Ed., "QUIC Loss Detection 1129 and Congestion Control", RFC 9002, DOI 10.17487/RFC9002, 1130 May 2021, . 1132 [RFC7540] Belshe, M., Peon, R., and M. Thomson, Ed., "Hypertext 1133 Transfer Protocol Version 2 (HTTP/2)", RFC 7540, 1134 DOI 10.17487/RFC7540, May 2015, 1135 . 1137 [RFC8081] Lilley, C., "The "font" Top-Level Media Type", RFC 8081, 1138 DOI 10.17487/RFC8081, February 2017, 1139 . 1141 Appendix A. Acknowledgements 1143 Roy Fielding presented the idea of using a header field for 1144 representing priorities in http://tools.ietf.org/agenda/83/slides/ 1145 slides-83-httpbis-5.pdf (http://tools.ietf.org/agenda/83/slides/ 1146 slides-83-httpbis-5.pdf). In https://github.com/pmeenan/http3- 1147 prioritization-proposal (https://github.com/pmeenan/http3- 1148 prioritization-proposal), Patrick Meenan advocated for representing 1149 the priorities using a tuple of urgency and concurrency. The ability 1150 to disable HTTP/2 prioritization is inspired by 1151 [I-D.lassey-priority-setting], authored by Brad Lassey and Lucas 1152 Pardue, with modifications based on feedback that was not 1153 incorporated into an update to that document. 1155 The motivation for defining an alternative to HTTP/2 priorities is 1156 drawn from discussion within the broad HTTP community. Special 1157 thanks to Roberto Peon, Martin Thomson and Netflix for text that was 1158 incorporated explicitly in this document. 1160 In addition to the people above, this document owes a lot to the 1161 extensive discussion in the HTTP priority design team, consisting of 1162 Alan Frindell, Andrew Galloni, Craig Taylor, Ian Swett, Kazuho Oku, 1163 Lucas Pardue, Matthew Cox, Mike Bishop, Roberto Peon, Robin Marx, Roy 1164 Fielding. 1166 Yang Chi contributed the section on retransmission scheduling. 1168 Appendix B. Change Log 1170 _RFC EDITOR: please remove this section before publication_ 1172 B.1. Since draft-ietf-httpbis-priority-08 1174 * Changelog fixups 1176 B.2. Since draft-ietf-httpbis-priority-07 1178 * Relax requirements of receiving SETTINGS_NO_RFC7540_PRIORITIES 1179 that changes value (#1714, #1725) 1181 * Clarify how intermediaries might use frames vs. headers (#1715, 1182 #1735) 1184 * Relax requirement when receiving a PRIORITY_UPDATE with an invalid 1185 structured field value (#1741, #1756) 1187 B.3. Since draft-ietf-httpbis-priority-06 1189 * Focus on editorial changes 1191 * Clarify rules about Sf-Dictionary handling in headers 1193 * Split policy for parameter IANA registry into two sections based 1194 on key length 1196 B.4. Since draft-ietf-httpbis-priority-05 1198 * Renamed SETTINGS_DEPRECATE_RFC7540_PRIORITIES to 1199 SETTINGS_NO_RFC7540_PRIORITIES 1201 * Clarify that senders of the HTTP/2 setting can use any alternative 1202 (#1679, #1705) 1204 B.5. Since draft-ietf-httpbis-priority-04 1206 * Renamed SETTINGS_DEPRECATE_HTTP2_PRIORITIES to 1207 SETTINGS_DEPRECATE_RFC7540_PRIORITIES (#1601) 1209 * Reoriented text towards RFC7540bis (#1561, #1601) 1211 * Clarify intermediary behavior (#1562) 1213 B.6. Since draft-ietf-httpbis-priority-03 1215 * Add statement about what this scheme applies to. Clarify 1216 extensions can use it but must define how themselves (#1550, 1217 #1559) 1219 * Describe scheduling considerations for the CONNECT method (#1495, 1220 #1544) 1222 * Describe scheduling considerations for retransmitted data (#1429, 1223 #1504) 1225 * Suggest intermediaries might avoid strict prioritization (#1562) 1227 B.7. Since draft-ietf-httpbis-priority-02 1229 * Describe considerations for server push prioritization (#1056, 1230 #1345) 1232 * Define HTTP/2 PRIORITY_UPDATE ID limits in HTTP/2 terms (#1261, 1233 #1344) 1235 * Add a Parameters registry (#1371) 1237 B.8. Since draft-ietf-httpbis-priority-01 1239 * PRIORITY_UPDATE frame changes (#1096, #1079, #1167, #1262, #1267, 1240 #1271) 1242 * Add section to describe server scheduling considerations (#1215, 1243 #1232, #1266) 1245 * Remove specific instructions related to intermediary fairness 1246 (#1022, #1264) 1248 B.9. Since draft-ietf-httpbis-priority-00 1250 * Move text around (#1217, #1218) 1251 * Editorial change to the default urgency. The value is 3, which 1252 was always the intent of previous changes. 1254 B.10. Since draft-kazuho-httpbis-priority-04 1256 * Minimize semantics of Urgency levels (#1023, #1026) 1258 * Reduce guidance about how intermediary implements merging priority 1259 signals (#1026) 1261 * Remove mention of CDN-Loop (#1062) 1263 * Editorial changes 1265 * Make changes due to WG adoption 1267 * Removed outdated Consideration (#118) 1269 B.11. Since draft-kazuho-httpbis-priority-03 1271 * Changed numbering from [-1,6] to [0,7] (#78) 1273 * Replaced priority scheme negotiation with HTTP/2 priority 1274 deprecation (#100) 1276 * Shorten parameter names (#108) 1278 * Expand on considerations (#105, #107, #109, #110, #111, #113) 1280 B.12. Since draft-kazuho-httpbis-priority-02 1282 * Consolidation of the problem statement (#61, #73) 1284 * Define SETTINGS_PRIORITIES for negotiation (#58, #69) 1286 * Define PRIORITY_UPDATE frame for HTTP/2 and HTTP/3 (#51) 1288 * Explain fairness issue and mitigations (#56) 1290 B.13. Since draft-kazuho-httpbis-priority-01 1292 * Explain how reprioritization might be supported. 1294 B.14. Since draft-kazuho-httpbis-priority-00 1296 * Expand urgency levels from 3 to 8. 1298 Authors' Addresses 1300 Kazuho Oku 1301 Fastly 1303 Email: kazuhooku@gmail.com 1305 Lucas Pardue 1306 Cloudflare 1308 Email: lucaspardue.24.7@gmail.com